Objectives: Most epidemiological studies of pesticides have used self-reports rather than quantitative measurements to assess exposures. The purpose of this study was to identify factors likely to affect exposure under actual field conditions and to measure the sensitivity and specificity of self-reported indications of exposure against urinary measures of herbicide exposure.
Methods: A sub-set of the participants in a retrospective cohort study of Ontario farm families volunteered for a pesticide exposure assessment study. Immediately prior, and subsequent to, handling the phenoxy-herbicides 2,4-dichlorophenoxyacetic acid (2,4-D) or 4-chloro-2-methylphenoxyacetic acid (MCPA) for the first time during the season, 126 pesticide applicators provided pre-exposure spot urine samples and a subsequent consecutive 24-h urine sample. At the same time, they completed a questionnaire on pesticide use and handling practices for the first day of pesticide application.
Results: Assuming that the presence of 2,4-D in the urine was a measure of true exposure and that questionnaire indications of 2,4-D use were the exposure classification subject to error, then the questionnaire's prediction of exposure had a sensitivity of 56.7% and specificity of 86.4%. The comparable values for MCPA were sensitivity and specificity of 91.6% and 67.4%, respectively. In multivariate models, the variables pesticide formulation, protective clothing/gear, application equipment, handling practice, and personal hygiene practice were significant as predictors of urinary herbicide levels in the first 24 h after application (or spraying) had been initiated (adjusted R(2)=44% for MCPA and 39% for 2,4-D).
Conclusions: Although similar domains of factors were associated with exposure in both models, the specific factors identified and the signs of the coefficients were sufficiently different between the final models for each herbicide that additional investigations appear to be warranted to determine the sources of the differences and assess the validity of the models and their ability to be generalised.